Try the calculator with your page counts or follow the worked example below.
Which page would you investigate first?
On 2 November our website still recorded 840 Bing AI citations that day. At the end of that week, one AEO roadmap accounted for nearly three-quarters of the citations in Bing’s page table. If we had reviewed only the site total, a strong week would have looked like evidence that the whole library was working. The page rows pointed to a different editorial decision: protect the roadmap and find out which buyer questions relied on it alone.
We used the native page table for 27 October–2 November 2026. It contained 6,474 citations across the listed pages. One AEO roadmap had 4,851, one GEO roadmap had 951, and the other listed pages together had 672. The site overview separately reported 6,742. Using the larger total would mix two reports with different coverage. AI Studio’s original Bing case record.

How do you calculate citation concentration?
Divide the citations attached to the largest page by the sum of citations in the same page table. For our AEO roadmap, 4,851 divided by 6,474 is 74.9%. Adding the GEO roadmap gives 5,802 divided by 6,474, or 89.6%. The second result is the share tied to the two pages together. It measures dependence within this export.
| Scenario | Citations remaining | Share of baseline |
|---|---|---|
| The largest page loses all its citations | 1,623 | 25.1% |
| Both leading pages lose half their combined citations | 3,573 | 55.2% |
| Both leading pages lose all their citations | 672 | 10.4% |
Each row holds the other pages constant. The rows are alternative scenarios for the same baseline. Bing later reported only 70 citations in the next week’s page table. Even removing both leading pages leaves 672 in the model. The further gap shows that a two-page explanation alone is incomplete. Coverage of the listed pages can also change between exports. The figures cannot tell us why Bing changed its sources.
Try the citation concentration calculator
Enter three counts from the same platform, property, report and date range. Start with the prefilled AI Studio figures or use your own. The calculator holds every other listed page constant.
Projected result: 672 citations would remain (10.4%). The two leading pages account for 89.6% of this page table.
How did the stress test compare with the actual decline?
In our case study the daily overview fell from 840 citations on 2 November to 48 on 3 November. The next two full weeks recorded 90 and 81 respectively. The daily overview can date the break. The weekly page table can identify exposure. Keeping both lets the editor decide when to investigate and which URLs to inspect.

How to use the result before rewriting a page
- Read the high-exposure page as a buyer. Identify the passage that answers a real purchasing question. Save the URL, exact passage, date and any answer capture that links to it. A high citation count alone cannot identify the sentence Bing used.
- Look for a genuinely different unanswered question. A roadmap can explain when to test. A separate review can show a manager what the figures mean when two agencies describe the same result differently. Give the second page its own evidence; a second version of the roadmap would add little to the buyer’s decision.
- Check the next comparable export. Record both the page-table total and the share of the two largest pages. A lower share can arise because a leading page disappeared or because new pages earned citations. Only the second explanation pairs wider coverage with retained performance. Inspect the URLs in both periods.
Which official reports can help you choose the next topic?
Microsoft’s Topics Portfolio beta groups grounding queries into themes for properties with enough data. Its opportunity list identifies queries in those themes where other sites appear. Citation volumes are ranges. A practical use is to shortlist one adjacent buyer question, open the existing result and ask what original evidence your team could supply. Microsoft describes this feature at the topic level. Our 89.6% figure came from the page table. Microsoft’s Topics Portfolio guide.
Google’s generative AI performance reports count impressions for URLs appearing in supported features. The November Search Analytics API update added reporting options including web multimodal activity and branded-query filters for eligible properties. Keep Google’s impression series alongside the Bing test with its own label and dates. Combining them into one “AI visibility” number would hide which platform actually changed. Google’s generative AI performance definitions · Google’s November API update.
When is a concentrated result acceptable?
A specialist with one authoritative product manual may reasonably want the manual to earn most citations. The useful question is what would become hard for a customer to find if it vanished. For AI Studio, two strategy roadmaps carried most of the listed citations while buyers also needed provider comparisons, reporting examples and finished work. That gave us a specific reason to invest in those formats. See the agency comparison and its evidence.
What can press coverage show about your brand?
Consider our food imagery. The Straits Times interviewed our founder and published the original burger photograph AI Studio used alongside our AI-enhanced version, crediting both images to AI Studio. A buyer can inspect that comparison on the newspaper’s domain. A Bing page export for aistudio.com.sg cannot count citations to the newspaper’s URL. To measure that route, open the actual AI answer and record the linked domain and the brand it describes. The existence of the story alone tells us nothing about whether an assistant used it. The Straits Times’ original report and photographs.
About the fieldwork
AI Studio’s editorial team calculated the scenarios from the public case documented by Carol Tan, founder of AI Studio. The case also records the historical rise from 620 citations during 24 April–23 June 2026 (61 days) to about 83,000 during 23 June–21 September 2026 (91 days). Inspect the dates and original dashboard capture.